Intel Merrifield SoC benchmark scores online

[Nemesis 1]

Heres a copy and pace were the z2580 uses 2 power planes

The Saltwell CPU core is still very similar to the first Intel Atom. Therefore, the performance per clock should be very similar (only very small enhancements). However, the CPU now features additional 256 KB low-power-cache (on a separate voltage plane) to allow the CPU and L2 cache to enter a lower power state. Furthermore, the Atom now supports some sort of Turbo Boost and can reach up to 1.8 GHz shortly (1.3 GHz base frequency). Furthermore, the CPU core can be downclocked to 100 MHz (in 100 MHz steps) to save power

 

[Nemesis 1]

It's pointless to try to evaluate a GPU based on the number of cores it has, or even worse based on whatever the manufacturer feels like calling cores. It's totally different from multi-core on a CPU.

Its pointless. Than why do we have these pre release topics

 

[MightyMalus]

Its likely the Z2580 . Google that and its really screwy. Some are saying its 14nm lol '

I did find 2 things of interest . There is a Z2580 L that goes in phones it 2 cpu cores and 2 gpu cores. The Z2580 W is for windows tablets and up . The funny part its 2cpu cores and 1 gpu core . Doesn't make sense unles its IB type gpu which does make sense.

Also in AT article on test the power draw of these atom and arm processors .
He had a hard time measuring the full power draw of qaulcomms cpu because of this

the CPU specific numbers for Qualcomm exclude any power consumed by the L2 cache. The total platform power numbers do include it however as they are measured at the battery.

The Z2580 has done the same thing as qualcomm

Was the Atom in the AT article 2core and HT capable? Don't remember.
One thing to note, the S4 chip was not a current chip, didn't have the Adreno 320. And also, the Tegra 3 on WinRT disables the power saver core it seems.
Pretty unfair comparison.
Hopefully they will test out phone chips soon and be more up-to date. And at least in Android.

 

[Nemesis 1]

They have tested phones and intel on compute has got tegra whipped but gpu not close . Nv for the win , When testing the only metric that will matter for the helper core won't chaange any scores other than battery life and in the phones intel is longer lasting battery life ON PHONES. The tegra 4 going against the Z2580 Intel will win compute and grapgics will go nv likely . battery life is going to be interesting , nv says tegra 4 is 30% better life at idle . We need reviews befor going any further . I don't expect hondo should be to interesting. Like might beat NV in compute MAYBE as for graphics its wait and see

 

[Homeles]

It's pointless to try to evaluate a GPU based on the number of cores it has, or even worse based on whatever the manufacturer feels like calling cores. It's totally different from multi-core on a CPU.

...huh? Assuming the architecture isn't going through a substantial change, it's the primary indicator of performance.

 

[Exophase]

...huh? Assuming the architecture isn't going through a substantial change, it's the primary indicator of performance.

The GPU is substantially different (SGX544 is Series 5XT which has major changes over SGX545 which is Series 5) but aren't you forgetting another major component of performance? You know, clock speed..

Saying that a single core GPU makes no sense for some design is totally arbitrary, especially if you say it's because it has two CPU cores.

 

[Nemesis 1]

You don't suppose that its IVB Igpu like ananda told was coming to intel 32nm I believe Anada said its 4eu. I really don't set it at 4eu at least 8 with the possiability to slice in another 8 just like G3 but smaller number of eus. Silver mont is all intel generation 7 igpu even the phone chip. So I guess we just wait and see what is on the Z2580 W . AT did say intel was going to do this on 32nm . Well the Z2580W would be the last chance for what Intel said it was going to do and what Ananda reported

 

[Exophase]

No of course the lower end Z2580 isn't going to have a Gen7 GPU with the higher end one having an SGX544MP2. I don't know you find an SGX544MP1 variant unfathomable. It's a big improvement vs SGX540 in current Medfield. I don't remember Anand ever saying that there'd be a 32nm Atom with Gen7. Do you have a link for that?

There has also never been any announcement or leak that Merrifield will use Gen7. On the Reddit AMA the Intel engineer said that they're going to catch up their IMG GPU offerings so they're no longer generations behind. That comment doesn't make a lot of sense if they're dropping it 100% in the near term.

 

[LogOver]

Same device in GLBenchmark:

http://www.glbenchmark.com/compare.jsp?D1=Intel+redhookbay&D2=Motorola+XT890&cols=2

 

[Nemesis 1]

Interesting. 2x /3x faster. Thats the z2580. If Silvermont improves graphics by as much . It will be a fine chip. Intel I believe said Silvermont will be 3x graphics and about 2x faster cpu than the z2580.

2x faster really not saying anything. it 4 cores against 2 so I would assume 2x faster. But than again intel has a tendency of going core against core . But they didn't do that with the z2580. They doubled up on everthing and thats the same figures 2x faster than medfield. Now if its 60%faster than the 4 threaded z2580 as it is also written . Than were talking some big gains. XT890 is no slow poke. So it was the motorola XT890 vs. the motorola 900. What I really like about the hand helds is intels referance designs . These are REAL products . Intel is serving notice . Will sell you our chips for your devices. Or well go it alone and sell our own devices. The Haswell referance design for ultras is really really nice. Intel needs to stop goofing off. And start selling there own branded devices. That referance Ultra intel showed at CES would be among the top sellers . Intel wants these people Lenovo and company To make ultras for under $1000. They don't seem to want to reduce rthere margins . Intel can force the issue by selling its own brand wth quality and value built in . Put it on the market for $799 and make the other players play.

Nazi grammer cops are such losers. Bill Clinton gave the game away and all the smart people failed to comprehend just exactly What it was that Billy boy did. Until you understand . How do you define is . The Judge understood as all judges have to LIE all the time as do lawyers . Law uses proper grammer. Find out what I am tring to tell you all here . Why was it when Billy boy said the depends on defining IS. THe JUdge stopped everthing ritht there and than . Than we get over to the house hereing and whats Billy Boy doing . He pulls a 1 dollar bill out of his pocket . and beginns slowly talking about the 1 dollar bill . and the writting on it . ERverthing came to a stop it was over . Billy threatened to tell the truth about our money and the TRUE grammer used BY law In a COURT of law . Its not a court of law . Its the port authority which is governened by no nation and uses true grammer , Look at a 1 dollar bill notice the square around the 1. find out what it means . Wake up .

 

[ShintaiDK]

 

[tangey]

my (reasonably informed) understanding is that Intel is sticking with IMG for graphics in smartphones for the next 2 gens at least.

Regarding the benchmark comparison linked above, is redhookbay for sure for smartphone or the one for tablet. I'm thinking its the one for tablets. I read quite a while back that the tablet one would be single core, and the smartphone would be dual core. The fill rate comparison would lead me to suggest that this is a single core 544. As I recall 544 gets 2 pixels per clock compared to 540. So theoritically at similar clocks ignoring all other things like bandwidth etc, 544 should get around x2 fill rate of 540. Those results would seem to be in the right ball park for that. For it to be a dual core 544, then they would have to be clocking way lower than the 540 in medfield to get that fill rate.

 

[Nemesis 1]

Reasonably informed. Z2580 is 2 544 Gpus its whats in the New Intel referance designed phone by intel . lenovo calls it the 900 series . They showed it at ces.

http://www.engadget.com/2012/08/29/intel-bay-trail-roadmap-leaked/

The chips should offer a burly horsepower bump over their predecessors, with up to four cores and clock speeds topping out at 2.4Ghz. The icing on the cake will be the integrated Gen 7 graphics engines of Ivy Bridge fame, featuring the same HD 4000 and HD 2500 GPU's as the grownup chips, but with only four "execution units"

 

[tangey]

I know what the Z2580 comprises. However is redhookbay an implementation on this, or is it a tablet formation. The glbenchmarks fill rate looks far too low vis-a-vis medfield to be 544mp2@533mhz which is what my expectations are for Z2580.

 

[Exophase]

I know what the Z2580 comprises. However is redhookbay an implementation on this, or is it a tablet formation. The glbenchmarks fill rate looks far too low vis-a-vis medfield to be 544mp2@533mhz which is what my expectations are for Z2580.

An SGX544 core has the same texture fillrate as an SGX540 when at the same clock (2 TMUs), and up to twice the ALU throughput and AFAIK stronger triangle setup, depth testing, and probably other improvements. So if the clocks are similar you'd expect 2-4x improvement, which is what you get in most of the relevant (offscreen) tests.

Anything external bandwidth limited won't see as much of an improvement vs Medfield. That could apply to the fillrate tests, particularly if they aren't using decent or any texture compression. 32bpp textures at that rate could easily hit bandwidth limits, depending on exactly what those are.

It scores close to A5X outside of synthetic tests, where A5X probably also has much higher peak bandwidth. That fits well for an SGX544MP2 at close to double the GPU clock speed A5X used. An SGX544MP1, as the phone variation is alleged to have (not sure if this applies to K900 or not), is out of the question for these results.

 

[tangey]

An SGX544 core has the same texture fillrate as an SGX540 when at the same clock (2 TMUs), and up to twice the ALU throughput and AFAIK stronger triangle setup, depth testing, and probably other improvements. So if the clocks are similar you'd expect 2-4x improvement, which is what you get in most of the relevant (offscreen) tests.

My bad on that one, was confusing medfield with 535 fill rate. Looked at again I suppose 700m fillrate for a 400mhz sgx540 isn't far from theoretical. Last I heard z2580 was supposed to clock 544mp2@533, in theory that should give it around 2g fill rate, instead of just over 1g. SGX seems to be able to get very close to it's theoritical max when every else is working good, all the apple socs hit very close to their theoritical fill rates. Ipad2 (543mp2@250mhz) gets around 1G fillrate, which is why I would be looking for a 2G fillrate on 544mp2@533.

But you are correct that other than fill rate, its performance is similar to the ipad3 in many offscreen tests.